How to Run an AI Training Needs Assessment

Most organisations buy AI training the way a nervous shopper buys a gym membership: they feel behind, they want to do something, and they pick a programme mostly to relieve the anxiety of not having one. Then they are surprised when it does not change much. The problem is almost never the training. It is that nobody ever established what the team actually needed before buying it.

A needs assessment is the unglamorous step that separates training spend that works from training spend that evaporates. It answers three questions: where is your team now, where does it need to be, and what is the specific gap that training has to close? Skip it, and you are buying a solution to a problem you have not defined.

This is a practical, step-by-step guide to running an AI training needs assessment — light enough that a small team can do it in a week, structured enough to give a larger organisation a defensible plan.

Why a Needs Assessment Matters More With AI

Training-needs analysis is not new; L&D teams have run it for decades. But AI raises the stakes in three specific ways that make skipping it especially costly.

First, the skill levels within a single team are wildly uneven. Unlike, say, a compliance course where everyone starts near zero, some of your people are already power users of AI while others have never opened a chatbot. Train them together without knowing this and you will bore half the room and lose the other half. Second, the highest-value use cases are function-specific and non-obvious — you cannot guess them from the outside, you have to go and find them. Third, the tools move so fast that generic "AI awareness" dates quickly; what your team needs is capability tied to their real work, which only an assessment reveals.

Buying AI training without a needs assessment is like prescribing medication without a diagnosis. It might help. It might do nothing. It might treat a condition the patient does not have — and you will not know which until the money is spent.

The Five-Step Assessment

Here is the process, in the order it should run. The whole thing scales — a small team can compress it into a few conversations; a large organisation will run each step more formally.

Step 1: Define the business outcome first

Before you look at skills at all, get clear on what the organisation is actually trying to achieve. Faster proposal turnaround? Lower cost of content production? Fewer hours lost to routine admin? Better client responsiveness? Training is a means, not an end, and if you cannot name the business outcome, you have no way to judge whether any training worked.

Write it down as a concrete, testable statement. "Improve AI skills" is not an outcome. "Cut the time to produce a first-draft client report from three hours to one" is. This single sentence will steer every other step of the assessment and every decision about what to buy.

Step 2: Map the current state honestly

Now find out where people actually are. You are gathering three things: current skill level, current tool usage, and current attitude. A short, anonymous survey works well for breadth — ask people to self-rate their confidence with AI, name any tools they already use, and describe their biggest concerns. Supplement it with a handful of one-to-one conversations for depth, because surveys flatten the nuance.

Pay special attention to the spread, not just the average. A team with a mean confidence of "moderate" might be evenly moderate, or it might be half experts and half beginners — and those two situations demand completely different training. Watch, too, for shadow AI: people quietly using unapproved tools. That tells you both where the appetite is and where the governance risk sits.

Step 3: Find the high-value use cases

This is the heart of the assessment and the step most often skipped. Sit with each function and identify the tasks that are (a) time-consuming, (b) repetitive or formulaic, and (c) important enough that speeding them up matters. These are where AI training will pay for itself.

The best way to surface them is to ask people to walk you through a typical week and flag the parts they find tedious. The tedium is the signal — the tasks people dread are very often exactly the ones AI is good at. You are looking to leave this step with a shortlist of specific, named workflows per function, not vague categories. "The team spends half a day each week reformatting the same weekly report" is a finding you can build training around. "Marketing could use AI" is not.

Step 4: Identify the gap and segment your people

You now have a target state (Step 1), a current state (Step 2), and the high-value use cases (Step 3). The gap is the difference between what people can do today and what they would need to be able to do to deliver those use cases. Name it specifically for each function.

Crucially, segment your people rather than treating them as one block. In most organisations they fall into three groups: a small set of advanced users who could become champions and coaches, a large middle who need structured, applied training to build real fluency, and a group of beginners or sceptics who need foundational literacy and some confidence-building first. Each segment needs a different intervention, and recognising this is what stops you buying one course for everyone and disappointing most of them. A structured AI competency framework is a useful tool for making this segmentation rigorous rather than impressionistic.

Step 5: Match the gap to the right training format

Only now — with outcome, current state, use cases, gap, and segments all in hand — are you ready to think about what to buy. And the assessment makes this decision far easier, because you know exactly what you are shopping for:

• If the gap is mostly foundational awareness across a broad group, an efficient generic literacy course covers it.
• If the gap is deep, workflow-specific capability in key functions, that is a case for bespoke training built around the use cases you identified.
• If leaders need to make strategic decisions about AI, that points to a dedicated business leaders programme rather than tool tutorials.
• If professionals need to become genuinely fluent in applying AI to their day-to-day work, an applied programme like AI for Professionals fits.

The point is that the format now follows from evidence rather than guesswork. Our guide to choosing the right AI training goes deeper on matching format to need once you have the assessment in hand.

Turning the Assessment Into a Plan

An assessment that sits in a document changes nothing. To make it useful, convert it into a short, concrete plan with four things in it:

1. The priority. You will find more gaps than you can address at once. Rank them by value and start with the one or two that most directly serve the business outcome from Step 1.
2. The intervention per segment. Spell out what each group gets — foundational, applied, or advanced — so nobody is over- or under-served.
3. The success measure. Define, before any training happens, how you will know it worked — tied to the business outcome and to behaviour, not to attendance. This is where a pre-training baseline earns its keep.
4. The reassessment point. Decide when you will re-run a lighter version of the assessment — a few months out — to see whether the gap actually closed and what the next priority should be.

Common Pitfalls to Avoid

A needs assessment can go wrong in ways that quietly undermine everything that follows. These are the traps we see most often.

Relying on self-assessment alone

Asking people to rate their own AI ability is useful, but it is not the whole picture. People are unreliable narrators of their own competence — some overrate themselves because they have used a chatbot a few times, others underrate themselves because they assume everyone else is further ahead. Triangulate. Pair the survey with a few real conversations and, where you can, a look at what people actually produce. The gap between what people say they can do and what they demonstrably do is often where the real training need hides.

Assessing skills without assessing tasks

It is tempting to treat the assessment as a pure skills audit — who can prompt, who understands the tools — and stop there. But skills in the abstract do not tell you what training to buy. The high-value use cases (Step 3) are what turn a skills gap into a training plan. An assessment that maps capability but never maps the actual work will produce generic conclusions and generic training, which is exactly the outcome you were trying to avoid.

Letting the loudest voice set the agenda

In any assessment, some people are more vocal than others. The enthusiast who loves AI and the sceptic who distrusts it will both make themselves heard, and neither should be allowed to define the need for everyone. Weight your findings by evidence and value, not by volume. The quiet team drowning in a repetitive task they never mention may represent a far bigger opportunity than the loud one already experimenting.

Treating the assessment as a one-time event

Perhaps the most important thing to understand about an AI needs assessment is that it goes stale. The tools change, people's skills move, and new use cases emerge. An assessment done once and filed away describes a moment that has already passed. Build in a lightweight reassessment — a shorter version run a few months later — so that your understanding of the gap keeps pace with reality. Capability-building is a loop, not a line.

The goal of a needs assessment is not a perfect document. It is a decision you can defend: this is where we are, this is where we need to be, and this is the specific training that closes the distance between the two.

Keep it proportionate

A needs assessment is meant to de-risk your investment, not become a project in itself. For a small team, the whole thing can be a survey, three conversations, and an afternoon of thinking — genuinely enough to avoid the biggest mistakes. For a large, multi-function organisation it warrants more rigour and probably outside help. Scale the effort to the size of the decision, but do not skip it entirely, however tempting the shortcut feels.

The organisations that get real value from AI training almost always did this work first. It is not glamorous, and it delays the satisfying moment of booking a programme by a week or two. But it is the difference between training that closes a gap you have actually measured and training that makes everyone feel productive while changing nothing. Diagnose first, then prescribe — and the training you buy will be the training you needed.